Hello @umaraziz
You’re correct that nodes getting stuck typically happens when transition conditions aren’t met(1). To help identify the issue, here are the key diagnostic steps:
Check Your Transition Conditions
Review the transition_condition.prompt field for your edges—the condition field is not used by the LLM to trigger transitions(2). Ensure your prompt is concise, declarative, and matches actual language patterns the user would say(2).
Common Solutions for Stuck Nodes
When node transitions aren’t working as expected(1):
- Review your transition conditions to ensure they precisely match your intended triggers(1)
- Add transition finetune examples to help the model understand your expectations(1)
- Make transition conditions more flexible and general(1)
- Add more nodes to cover edge cases, particularly global nodes for handling unexpected situations(1)
For Dynamic Variable Issues
If you’re using dynamic variables in your transition conditions, the node may be waiting for a user response before it can transition(3). The variable needs to be populated and the user needs to provide input for the transition to evaluate(3).
Debugging Steps
You can find node transitions inside the call transcript in the history tab, which will show the node names that it transitions from and to(2). This helps identify exactly when transitions occur or fail(2).
If Instructions Aren’t Being Followed
If the agent isn’t following instructions within a node before transitioning(1):
- Split the node into multiple nodes for better consistency(1)
- Change the node model to a more capable LLM(1)
- Add conversation finetune examples(1)
- Adjust the LLM temperature if responses are inconsistent(1)
To provide more specific guidance, share the call ID and describe the expected behavior at the specific timestamp where the node gets stuck.